3D Reconstruction Method Research Based On The Video Sequence Of Complex

Along with the computer vision technology in application of various fields, all kinds of information based on the three dimensions are more and more causing the attention of people. From single perspective view to more interactive browsing, three-dimensional visual enjoyment of high lifelike has gradually became people’s needs, the way of percepting information also has undergone great changes.3D reconstruction of space target is an important content in the research of computer vision. Due to the image and video information of quality is uneven in real life, some low quality images or videos increase the error in the process of3D reconstruction, they must be removed. In addition, images with close shooting angle or continuous video frames become redundant data, due to the short baseline between them, they are little contribution to the reconstruction, even increase the error of reconstruction, so they must also be removed. Therefore, this article puts forward the key frames feature extraction algorithm, which is proposed to overcome the problems mentioned above. The main content of this paper is as follows:1. In this paper, key frame extraction algorithm is put forward in the multi-view video sequence in the3D scene reconstruction effectively based on computer vision, the algorithm is mainly based on color feature, shape feature and basic image invariant feature points, which generates a multi-dimensional vector as constraint conditions to delete a lot of redundant information, accordingly improves the running time and matching accuracy.2. This paper puts forward an algorithm which base on convection matching of stereo matching algorithm. This algorithm first introduces inhibiting factor to eliminate the Harris feature points of accumulation phenomenon, and then using Sift feature descriptor based matching, the random sampling algorithm is used to estimate fundamental matrix, and at the same time eliminate false matching, finally on the basis of estimating matrix guide the convection matching.Finally, in this paper, the complex environment of the target objects in video sequences has realized the3D reconstruction, this paper demonstrates that the proposed method is feasible, then the result of the experiment are analyzed. Method is proposed to solve the image fuzzy or shooting angle near and video information of redundancy, which has good effect on such issues. Experiments also show that the algorithm can well reproduce the object structure, which is an effective3D reconstruction of stereo matching algorithm.